The agent firewall.
See, control or block what any AI agent can access. Shadow or approved.
A Runtime Enforcement Layer. Real-time observability and response for AI applications and agents across the enterprise. Set what each agent can read, write and reach. Ospiri enforces access controls inline, at the endpoint.
Guardrails for AI, without stalling critical business workflows.
Ospiri adds a purpose-built, context-aware security layer for agents on the endpoint, designed to run alongside your EDR and SOC tooling.
| The question | Typical EDR or AI gateway | With Ospiri |
|---|---|---|
| Can I restrict which sensitive files my Claude and Copilot agents open? | Not per agent | Yes. Inline access policy, with the safe copy for changes. |
| Can I see what an agent's skills are built to do? | Not typically | Yes. Skills are analyzed and reported on, including what they are built to do. |
| Can you stop one action without stopping the agent? | Partially. Only blocks for unauthorized AI tools. | Yes. Block it, hold it for approval, or send it to a safe copy. |
| Where is each agent allowed to read, write and connect? | Varies by vendor add-on | Set per agent, per team and per type of data |
| Does the policy carry over if you change EDR? | No | Yes. Vendor agnostic: the same policy beside any EDR. |
From the org chart down to the file an agent tried to open.
Every agent is mapped to the person and team running it. You see whether it was approved, what it can reach, and what Ospiri did on the endpoint.
of unauthorized AI agent transactions through 2028 will come from internal violations of enterprise policy (oversharing, unacceptable use, misguided AI behavior) rather than from malicious attacks.
Gartner, Market Guide for Guardian Agents, G00836388, 25 Feb 2026
Finding agents is becoming a standard feature. The hard part is containing an agent without stopping the work. That is what Ospiri is built for.
Three situations we hear about most.
“Copilot is going out company-wide. Leadership wants one view of every agent.”
Use case · Large enterprise on Microsoft
“Agent rollout was on hold. It is back on, with access set by team.”
Use case · Financial services
“Block the note-takers and app builders you have not approved. Keep the ones you have.”
Use case · Software company
A minor violation shouldn't stop critical work.
Most controls can only allow an action or kill the agent. Ospiri has a third option. If an agent tries to change something outside its limits, the change goes to a safe copy instead of the real file. The agent keeps working, and the original stays exactly as it was.
edits config.ts
SDLC guardrails
A coding agent edits a file outside its repository. The change lands in the safe copy; the original is untouched.
Access controls by group
Agents used by sales can't touch the engineering file share or the finance database.
Note-taking apps
A newly installed note-taker or AI app starts contained until someone approves it.
Restricting where agents can write
Allow an agent to read a folder but not change it, or keep it away from a credential store entirely.
Close the agent gap.
The agent gap is the distance between the agents you approved and what runs on the endpoint. Ospiri closes it in four steps. The first two are useful before you enforce anything.
Presence
Which agents are running or sitting dormant, on which endpoints, under whose account. Personal accounts on managed laptops included.
ollama (llama3), 2 MCP servers
Access
What each one can reach: files, credentials, MCP servers, internal systems.
github token, jira MCP
Action
What it actually did, attributed to the agent rather than the person it ran as.
config.ts outside its repo
Control
What it may do next. Enforced on the endpoint, before the action completes.
unreviewed skill → block
Close the gap and you can approve more agents, sooner.
CISO
Putting an AI risk framework in place, and running it.
Security engineering
Can we turn on enforcement without breaking a legitimate build?
Security operations
What happened, in what order, and was it the person or the agent?
IT and endpoint operations
What does this cost my endpoints, and does it work with the EDR we run?
Compliance
Audits and reporting on agent activity.
Board and executive team
One view of AI governance across the company.
Do you know what AI is running on your endpoints now?
A lightweight scanner lists the agents, skills and MCP servers on the endpoints you choose. We walk through the findings with you, and the scanner is removed when the week ends.
Add control later, on the machines where it matters.